Complex SQL query by pattern with timestamps - sql

Question

I have the following info in my SQLite database:
ID | timestamp | val
1 | 1577644027 | 0
2 | 1577644028 | 0
3 | 1577644029 | 1
4 | 1577644030 | 1
5 | 1577644031 | 2
6 | 1577644032 | 2
7 | 1577644033 | 3
8 | 1577644034 | 2
9 | 1577644035 | 1
10 | 1577644036 | 0
11 | 1577644037 | 1
12 | 1577644038 | 1
13 | 1577644039 | 1
14 | 1577644040 | 0
I want to perform a query that returns the elements that compose an episode. An episode is a set of ordered registers that comply the following requirements:
The first element is greater than zero.
The previous element of the first one is zero.
The last element is greater than zero.
The next element of the last one is zero.
The expected result of the query on this example would be something like this:
[
[{"id":3, tmstamp:1577644029, value:1}
{"id":4, tmstamp:1577644030, value:1}
{"id":5, tmstamp:1577644031, value:2}
{"id":6, tmstamp:1577644032, value:2}
{"id":7, tmstamp:1577644033, value:3}
{"id":8, tmstamp:1577644034, value:2}
{"id":9, tmstamp:1577644035, value:1}],
[{"id":11, tmstamp:1577644037, value:1}
{"id":12, tmstamp:1577644038, value:1}
{"id":13, tmstamp:1577644039, value:1}]
]
Currently, I am avoiding this query and I am using an auxiliary table to store the initial and end timestamp of episodes, but this is only because I do not know how to perform this query.
Threfore, my question is quite straightforward: does anyone know how can I perform this query in order to obtain something similar to the stated ouput?

Answer 1

This answer assumes that the "before" and "after" conditions are not really important. That is, an episode can be the first row in the table.
You can identify the episodes by counting the number of 0s before each row. Then filter out the 0 values:
select t.*,
dense_rank() over (order by grp) as episode
from (select t.*,
sum(case when val = 0 then 1 else 0 end) over (order by timestamp) as grp
from t
) t
where val <> 0;
If this is not the case, then lag() and lead() and a cumulative sum can handle the previous value being 0:
select t.*,
sum(case when prev_val = 0 and val > 0 then 1 else 0 end) over (order by timestamp) as episode
from (select t.*,
lag(val) over (order by timestamp) as prev_val,
lead(val) over (order by timestamp) as next_val
from t
) t
where val <> 0;

Answer 2

If you want the result as JSON objects then you must use the JSON1 Extension functions of SQLite:
with cte as (
select *, sum(val = 0) over (order by timestamp) grp
from tablename
)
select
json_group_array(
json_object('id', id, 'timestamp', timestamp, 'val', val)
) result
from cte
where val > 0
group by grp
See the demo.
Results:
| result |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [{"id":3,"timestamp":1577644029,"val":1},{"id":4,"timestamp":1577644030,"val":1},{"id":5,"timestamp":1577644031,"val":2},{"id":6,"timestamp":1577644032,"val":2},{"id":7,"timestamp":1577644033,"val":3},{"id":8,"timestamp":1577644034,"val":2},{"id":9,"timestamp":1577644035,"val":1}] |
| [{"id":11,"timestamp":1577644037,"val":1},{"id":12,"timestamp":1577644038,"val":1},{"id":13,"timestamp":1577644039,"val":1}] |